The tweet was deleted by the author.
But we saved everything 🙂.
Karim Atiyeh announced an innovation in artificial intelligence with PorTAL, a framework that separates task representation from a model's weights. Traditionally, fine-tuning requires training within a single model, making transfer between models cumbersome. PorTAL addresses this by extracting the learned task encoding, enabling users to port tasks to new models by simply fitting a thin converter, removing the need for full retraining. This approach could streamline workflows for organizations working with multiple AI systems.
Atiyeh recently announced a formal leadership change focused on technology and systems design at the company. Earlier, his company expanded to serve more than 70,000 businesses, emphasizing AI-driven finance at $44 billion in volume. These developments set the stage for the introduction of PorTAL.